src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRF_VdwLJ_GeomW3P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW3P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwLJ_GeomW3P1_VF_avx_128_fma_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  90     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  98     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  99     __m128d          dummy_mask,cutoff_mask;
 100     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 101     __m128d          one     = _mm_set1_pd(1.0);
 102     __m128d          two     = _mm_set1_pd(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm_set1_pd(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116     krf              = _mm_set1_pd(fr->ic->k_rf);
 117     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 118     crf              = _mm_set1_pd(fr->ic->c_rf);
 119     nvdwtype         = fr->ntype;
 120     vdwparam         = fr->nbfp;
 121     vdwtype          = mdatoms->typeA;
 122
 123     /* Setup water-specific parameters */
 124     inr              = nlist->iinr[0];
 125     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 126     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 127     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 128     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134
 135     outeriter        = 0;
 136     inneriter        = 0;
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 154                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 155
 156         fix0             = _mm_setzero_pd();
 157         fiy0             = _mm_setzero_pd();
 158         fiz0             = _mm_setzero_pd();
 159         fix1             = _mm_setzero_pd();
 160         fiy1             = _mm_setzero_pd();
 161         fiz1             = _mm_setzero_pd();
 162         fix2             = _mm_setzero_pd();
 163         fiy2             = _mm_setzero_pd();
 164         fiz2             = _mm_setzero_pd();
 165
 166         /* Reset potential sums */
 167         velecsum         = _mm_setzero_pd();
 168         vvdwsum          = _mm_setzero_pd();
 169
 170         /* Start inner kernel loop */
 171         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 172         {
 173
 174             /* Get j neighbor index, and coordinate index */
 175             jnrA             = jjnr[jidx];
 176             jnrB             = jjnr[jidx+1];
 177             j_coord_offsetA  = DIM*jnrA;
 178             j_coord_offsetB  = DIM*jnrB;
 179
 180             /* load j atom coordinates */
 181             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 182                                               &jx0,&jy0,&jz0);
 183
 184             /* Calculate displacement vector */
 185             dx00             = _mm_sub_pd(ix0,jx0);
 186             dy00             = _mm_sub_pd(iy0,jy0);
 187             dz00             = _mm_sub_pd(iz0,jz0);
 188             dx10             = _mm_sub_pd(ix1,jx0);
 189             dy10             = _mm_sub_pd(iy1,jy0);
 190             dz10             = _mm_sub_pd(iz1,jz0);
 191             dx20             = _mm_sub_pd(ix2,jx0);
 192             dy20             = _mm_sub_pd(iy2,jy0);
 193             dz20             = _mm_sub_pd(iz2,jz0);
 194
 195             /* Calculate squared distance and things based on it */
 196             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 197             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 198             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 199
 200             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 201             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 202             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 203
 204             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 205             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 206             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 207
 208             /* Load parameters for j particles */
 209             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 210             vdwjidx0A        = 2*vdwtype[jnrA+0];
 211             vdwjidx0B        = 2*vdwtype[jnrB+0];
 212
 213             fjx0             = _mm_setzero_pd();
 214             fjy0             = _mm_setzero_pd();
 215             fjz0             = _mm_setzero_pd();
 216
 217             /**************************
 218              * CALCULATE INTERACTIONS *
 219              **************************/
 220
 221             /* Compute parameters for interactions between i and j atoms */
 222             qq00             = _mm_mul_pd(iq0,jq0);
 223             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 224                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 225
 226             /* REACTION-FIELD ELECTROSTATICS */
 227             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 228             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 229
 230             /* LENNARD-JONES DISPERSION/REPULSION */
 231
 232             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 233             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 234             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 235             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 236             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 237
 238             /* Update potential sum for this i atom from the interaction with this j atom. */
 239             velecsum         = _mm_add_pd(velecsum,velec);
 240             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 241
 242             fscal            = _mm_add_pd(felec,fvdw);
 243
 244             /* Update vectorial force */
 245             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 246             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 247             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 248
 249             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 250             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 251             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 252
 253             /**************************
 254              * CALCULATE INTERACTIONS *
 255              **************************/
 256
 257             /* Compute parameters for interactions between i and j atoms */
 258             qq10             = _mm_mul_pd(iq1,jq0);
 259
 260             /* REACTION-FIELD ELECTROSTATICS */
 261             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 262             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 263
 264             /* Update potential sum for this i atom from the interaction with this j atom. */
 265             velecsum         = _mm_add_pd(velecsum,velec);
 266
 267             fscal            = felec;
 268
 269             /* Update vectorial force */
 270             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 271             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 272             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 273
 274             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 275             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 276             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 277
 278             /**************************
 279              * CALCULATE INTERACTIONS *
 280              **************************/
 281
 282             /* Compute parameters for interactions between i and j atoms */
 283             qq20             = _mm_mul_pd(iq2,jq0);
 284
 285             /* REACTION-FIELD ELECTROSTATICS */
 286             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 287             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 288
 289             /* Update potential sum for this i atom from the interaction with this j atom. */
 290             velecsum         = _mm_add_pd(velecsum,velec);
 291
 292             fscal            = felec;
 293
 294             /* Update vectorial force */
 295             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 296             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 297             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 298
 299             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 300             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 301             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 302
 303             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 304
 305             /* Inner loop uses 120 flops */
 306         }
 307
 308         if(jidx<j_index_end)
 309         {
 310
 311             jnrA             = jjnr[jidx];
 312             j_coord_offsetA  = DIM*jnrA;
 313
 314             /* load j atom coordinates */
 315             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 316                                               &jx0,&jy0,&jz0);
 317
 318             /* Calculate displacement vector */
 319             dx00             = _mm_sub_pd(ix0,jx0);
 320             dy00             = _mm_sub_pd(iy0,jy0);
 321             dz00             = _mm_sub_pd(iz0,jz0);
 322             dx10             = _mm_sub_pd(ix1,jx0);
 323             dy10             = _mm_sub_pd(iy1,jy0);
 324             dz10             = _mm_sub_pd(iz1,jz0);
 325             dx20             = _mm_sub_pd(ix2,jx0);
 326             dy20             = _mm_sub_pd(iy2,jy0);
 327             dz20             = _mm_sub_pd(iz2,jz0);
 328
 329             /* Calculate squared distance and things based on it */
 330             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 331             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 332             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 333
 334             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 335             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 336             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 337
 338             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 339             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 340             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 341
 342             /* Load parameters for j particles */
 343             jq0              = _mm_load_sd(charge+jnrA+0);
 344             vdwjidx0A        = 2*vdwtype[jnrA+0];
 345
 346             fjx0             = _mm_setzero_pd();
 347             fjy0             = _mm_setzero_pd();
 348             fjz0             = _mm_setzero_pd();
 349
 350             /**************************
 351              * CALCULATE INTERACTIONS *
 352              **************************/
 353
 354             /* Compute parameters for interactions between i and j atoms */
 355             qq00             = _mm_mul_pd(iq0,jq0);
 356             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 357
 358             /* REACTION-FIELD ELECTROSTATICS */
 359             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 360             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 361
 362             /* LENNARD-JONES DISPERSION/REPULSION */
 363
 364             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 365             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 366             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 367             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 368             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 369
 370             /* Update potential sum for this i atom from the interaction with this j atom. */
 371             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 372             velecsum         = _mm_add_pd(velecsum,velec);
 373             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 374             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 375
 376             fscal            = _mm_add_pd(felec,fvdw);
 377
 378             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 379
 380             /* Update vectorial force */
 381             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 382             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 383             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 384
 385             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 386             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 387             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 388
 389             /**************************
 390              * CALCULATE INTERACTIONS *
 391              **************************/
 392
 393             /* Compute parameters for interactions between i and j atoms */
 394             qq10             = _mm_mul_pd(iq1,jq0);
 395
 396             /* REACTION-FIELD ELECTROSTATICS */
 397             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 398             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 399
 400             /* Update potential sum for this i atom from the interaction with this j atom. */
 401             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 402             velecsum         = _mm_add_pd(velecsum,velec);
 403
 404             fscal            = felec;
 405
 406             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 407
 408             /* Update vectorial force */
 409             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 410             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 411             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 412
 413             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 414             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 415             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 416
 417             /**************************
 418              * CALCULATE INTERACTIONS *
 419              **************************/
 420
 421             /* Compute parameters for interactions between i and j atoms */
 422             qq20             = _mm_mul_pd(iq2,jq0);
 423
 424             /* REACTION-FIELD ELECTROSTATICS */
 425             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 426             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 427
 428             /* Update potential sum for this i atom from the interaction with this j atom. */
 429             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 430             velecsum         = _mm_add_pd(velecsum,velec);
 431
 432             fscal            = felec;
 433
 434             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 435
 436             /* Update vectorial force */
 437             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 438             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 439             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 440
 441             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 442             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 443             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 444
 445             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 446
 447             /* Inner loop uses 120 flops */
 448         }
 449
 450         /* End of innermost loop */
 451
 452         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 453                                               f+i_coord_offset,fshift+i_shift_offset);
 454
 455         ggid                        = gid[iidx];
 456         /* Update potential energies */
 457         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 458         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 459
 460         /* Increment number of inner iterations */
 461         inneriter                  += j_index_end - j_index_start;
 462
 463         /* Outer loop uses 20 flops */
 464     }
 465
 466     /* Increment number of outer iterations */
 467     outeriter        += nri;
 468
 469     /* Update outer/inner flops */
 470
 471     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*120);
 472 }
 473 /*
 474  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW3P1_F_avx_128_fma_double
 475  * Electrostatics interaction: ReactionField
 476  * VdW interaction:            LennardJones
 477  * Geometry:                   Water3-Particle
 478  * Calculate force/pot:        Force
 479  */
 480 void
 481 nb_kernel_ElecRF_VdwLJ_GeomW3P1_F_avx_128_fma_double
 482                     (t_nblist                    * gmx_restrict       nlist,
 483                      rvec                        * gmx_restrict          xx,
 484                      rvec                        * gmx_restrict          ff,
 485                      t_forcerec                  * gmx_restrict          fr,
 486                      t_mdatoms                   * gmx_restrict     mdatoms,
 487                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 488                      t_nrnb                      * gmx_restrict        nrnb)
 489 {
 490     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 491      * just 0 for non-waters.
 492      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 493      * jnr indices corresponding to data put in the four positions in the SIMD register.
 494      */
 495     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 496     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 497     int              jnrA,jnrB;
 498     int              j_coord_offsetA,j_coord_offsetB;
 499     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 500     real             rcutoff_scalar;
 501     real             *shiftvec,*fshift,*x,*f;
 502     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 503     int              vdwioffset0;
 504     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 505     int              vdwioffset1;
 506     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 507     int              vdwioffset2;
 508     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 509     int              vdwjidx0A,vdwjidx0B;
 510     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 511     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 512     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 513     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 514     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 515     real             *charge;
 516     int              nvdwtype;
 517     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 518     int              *vdwtype;
 519     real             *vdwparam;
 520     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 521     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 522     __m128d          dummy_mask,cutoff_mask;
 523     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 524     __m128d          one     = _mm_set1_pd(1.0);
 525     __m128d          two     = _mm_set1_pd(2.0);
 526     x                = xx[0];
 527     f                = ff[0];
 528
 529     nri              = nlist->nri;
 530     iinr             = nlist->iinr;
 531     jindex           = nlist->jindex;
 532     jjnr             = nlist->jjnr;
 533     shiftidx         = nlist->shift;
 534     gid              = nlist->gid;
 535     shiftvec         = fr->shift_vec[0];
 536     fshift           = fr->fshift[0];
 537     facel            = _mm_set1_pd(fr->epsfac);
 538     charge           = mdatoms->chargeA;
 539     krf              = _mm_set1_pd(fr->ic->k_rf);
 540     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 541     crf              = _mm_set1_pd(fr->ic->c_rf);
 542     nvdwtype         = fr->ntype;
 543     vdwparam         = fr->nbfp;
 544     vdwtype          = mdatoms->typeA;
 545
 546     /* Setup water-specific parameters */
 547     inr              = nlist->iinr[0];
 548     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 549     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 550     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 551     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 552
 553     /* Avoid stupid compiler warnings */
 554     jnrA = jnrB = 0;
 555     j_coord_offsetA = 0;
 556     j_coord_offsetB = 0;
 557
 558     outeriter        = 0;
 559     inneriter        = 0;
 560
 561     /* Start outer loop over neighborlists */
 562     for(iidx=0; iidx<nri; iidx++)
 563     {
 564         /* Load shift vector for this list */
 565         i_shift_offset   = DIM*shiftidx[iidx];
 566
 567         /* Load limits for loop over neighbors */
 568         j_index_start    = jindex[iidx];
 569         j_index_end      = jindex[iidx+1];
 570
 571         /* Get outer coordinate index */
 572         inr              = iinr[iidx];
 573         i_coord_offset   = DIM*inr;
 574
 575         /* Load i particle coords and add shift vector */
 576         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 577                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 578
 579         fix0             = _mm_setzero_pd();
 580         fiy0             = _mm_setzero_pd();
 581         fiz0             = _mm_setzero_pd();
 582         fix1             = _mm_setzero_pd();
 583         fiy1             = _mm_setzero_pd();
 584         fiz1             = _mm_setzero_pd();
 585         fix2             = _mm_setzero_pd();
 586         fiy2             = _mm_setzero_pd();
 587         fiz2             = _mm_setzero_pd();
 588
 589         /* Start inner kernel loop */
 590         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 591         {
 592
 593             /* Get j neighbor index, and coordinate index */
 594             jnrA             = jjnr[jidx];
 595             jnrB             = jjnr[jidx+1];
 596             j_coord_offsetA  = DIM*jnrA;
 597             j_coord_offsetB  = DIM*jnrB;
 598
 599             /* load j atom coordinates */
 600             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 601                                               &jx0,&jy0,&jz0);
 602
 603             /* Calculate displacement vector */
 604             dx00             = _mm_sub_pd(ix0,jx0);
 605             dy00             = _mm_sub_pd(iy0,jy0);
 606             dz00             = _mm_sub_pd(iz0,jz0);
 607             dx10             = _mm_sub_pd(ix1,jx0);
 608             dy10             = _mm_sub_pd(iy1,jy0);
 609             dz10             = _mm_sub_pd(iz1,jz0);
 610             dx20             = _mm_sub_pd(ix2,jx0);
 611             dy20             = _mm_sub_pd(iy2,jy0);
 612             dz20             = _mm_sub_pd(iz2,jz0);
 613
 614             /* Calculate squared distance and things based on it */
 615             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 616             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 617             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 618
 619             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 620             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 621             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 622
 623             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 624             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 625             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 626
 627             /* Load parameters for j particles */
 628             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 629             vdwjidx0A        = 2*vdwtype[jnrA+0];
 630             vdwjidx0B        = 2*vdwtype[jnrB+0];
 631
 632             fjx0             = _mm_setzero_pd();
 633             fjy0             = _mm_setzero_pd();
 634             fjz0             = _mm_setzero_pd();
 635
 636             /**************************
 637              * CALCULATE INTERACTIONS *
 638              **************************/
 639
 640             /* Compute parameters for interactions between i and j atoms */
 641             qq00             = _mm_mul_pd(iq0,jq0);
 642             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 643                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 644
 645             /* REACTION-FIELD ELECTROSTATICS */
 646             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 647
 648             /* LENNARD-JONES DISPERSION/REPULSION */
 649
 650             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 651             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 652
 653             fscal            = _mm_add_pd(felec,fvdw);
 654
 655             /* Update vectorial force */
 656             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 657             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 658             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 659
 660             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 661             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 662             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 663
 664             /**************************
 665              * CALCULATE INTERACTIONS *
 666              **************************/
 667
 668             /* Compute parameters for interactions between i and j atoms */
 669             qq10             = _mm_mul_pd(iq1,jq0);
 670
 671             /* REACTION-FIELD ELECTROSTATICS */
 672             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 673
 674             fscal            = felec;
 675
 676             /* Update vectorial force */
 677             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 678             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 679             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 680
 681             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 682             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 683             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 684
 685             /**************************
 686              * CALCULATE INTERACTIONS *
 687              **************************/
 688
 689             /* Compute parameters for interactions between i and j atoms */
 690             qq20             = _mm_mul_pd(iq2,jq0);
 691
 692             /* REACTION-FIELD ELECTROSTATICS */
 693             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 694
 695             fscal            = felec;
 696
 697             /* Update vectorial force */
 698             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 699             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 700             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 701
 702             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 703             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 704             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 705
 706             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 707
 708             /* Inner loop uses 100 flops */
 709         }
 710
 711         if(jidx<j_index_end)
 712         {
 713
 714             jnrA             = jjnr[jidx];
 715             j_coord_offsetA  = DIM*jnrA;
 716
 717             /* load j atom coordinates */
 718             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 719                                               &jx0,&jy0,&jz0);
 720
 721             /* Calculate displacement vector */
 722             dx00             = _mm_sub_pd(ix0,jx0);
 723             dy00             = _mm_sub_pd(iy0,jy0);
 724             dz00             = _mm_sub_pd(iz0,jz0);
 725             dx10             = _mm_sub_pd(ix1,jx0);
 726             dy10             = _mm_sub_pd(iy1,jy0);
 727             dz10             = _mm_sub_pd(iz1,jz0);
 728             dx20             = _mm_sub_pd(ix2,jx0);
 729             dy20             = _mm_sub_pd(iy2,jy0);
 730             dz20             = _mm_sub_pd(iz2,jz0);
 731
 732             /* Calculate squared distance and things based on it */
 733             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 734             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 735             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 736
 737             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 738             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 739             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 740
 741             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 742             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 743             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 744
 745             /* Load parameters for j particles */
 746             jq0              = _mm_load_sd(charge+jnrA+0);
 747             vdwjidx0A        = 2*vdwtype[jnrA+0];
 748
 749             fjx0             = _mm_setzero_pd();
 750             fjy0             = _mm_setzero_pd();
 751             fjz0             = _mm_setzero_pd();
 752
 753             /**************************
 754              * CALCULATE INTERACTIONS *
 755              **************************/
 756
 757             /* Compute parameters for interactions between i and j atoms */
 758             qq00             = _mm_mul_pd(iq0,jq0);
 759             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 760
 761             /* REACTION-FIELD ELECTROSTATICS */
 762             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 763
 764             /* LENNARD-JONES DISPERSION/REPULSION */
 765
 766             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 767             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 768
 769             fscal            = _mm_add_pd(felec,fvdw);
 770
 771             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 772
 773             /* Update vectorial force */
 774             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 775             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 776             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 777
 778             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 779             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 780             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 781
 782             /**************************
 783              * CALCULATE INTERACTIONS *
 784              **************************/
 785
 786             /* Compute parameters for interactions between i and j atoms */
 787             qq10             = _mm_mul_pd(iq1,jq0);
 788
 789             /* REACTION-FIELD ELECTROSTATICS */
 790             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 791
 792             fscal            = felec;
 793
 794             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 795
 796             /* Update vectorial force */
 797             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 798             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 799             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 800
 801             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 802             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 803             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 804
 805             /**************************
 806              * CALCULATE INTERACTIONS *
 807              **************************/
 808
 809             /* Compute parameters for interactions between i and j atoms */
 810             qq20             = _mm_mul_pd(iq2,jq0);
 811
 812             /* REACTION-FIELD ELECTROSTATICS */
 813             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 814
 815             fscal            = felec;
 816
 817             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 818
 819             /* Update vectorial force */
 820             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 821             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 822             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 823
 824             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 825             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 826             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 827
 828             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 829
 830             /* Inner loop uses 100 flops */
 831         }
 832
 833         /* End of innermost loop */
 834
 835         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 836                                               f+i_coord_offset,fshift+i_shift_offset);
 837
 838         /* Increment number of inner iterations */
 839         inneriter                  += j_index_end - j_index_start;
 840
 841         /* Outer loop uses 18 flops */
 842     }
 843
 844     /* Increment number of outer iterations */
 845     outeriter        += nri;
 846
 847     /* Update outer/inner flops */
 848
 849     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*100);
 850 }